Global fabless semiconductor companies and tech giants are expected to adopt the 3-nanometer process as their mainstream technology this year. As most companies are likely to place their orders with Taiwan’s TSMC, the world’s biggest contract chipmaker, the market share gap between TSMC and South Korea’s Samsung Electronics is projected to widen further. Industry insiders say Samsung Electronic’s contract manufacturing division’s prototypes have lagged behind TSMC’s in terms of yield and power efficiency, costing it an early lead in the 3nm race.
Seven companies, including Nvidia, AMD, Intel, Qualcomm, MediaTek, Apple and Google, have decided to use TSMC’s 3nm process, according to industry reports on June 17. Google and Qualcomm, which Samsung Electronics’ foundry division had been working hard to secure for next-generation chip volumes, ultimately chose TSMC after careful consideration.
Google, which had outsourced its Tensor processor to Samsung’s foundry division until the fourth generation, will switch to TSMC’s fabs for the fifth generation, where the 3nm process will be introduced. Similarly, Qualcomm’s Snapdragon 8 Gen 4 products, which Samsung foundry has strived to secure since last year, will also be outsourced to TSMC for initial production.
Despite entering its third year since announcing the mass production of 3nm process chips, Samsung has struggled to secure customers. Samsung was the first chipmaker to begin mass production of the 3nm process using the gate-all-around field-effect transistor (GAAFET) process in June 2022. However, its first-generation 3nm process, the N3 node (SF3E), has only been adopted in niche markets such as cryptocurrency mining chips due to its worse-than-expected yield and performance. Samsung’s System LSI division’s Exynos 2500, developed in-house and produced using Samsung Foundry’s 3nm process, has faced similar yield issues.
Experts point to low yield and power efficiency as the biggest issues plaguing Samsung’s 3nm process. Although Samsung has been working on controlling power consumption and heat generation - key factors in advanced semiconductor processes - the efficiency of the chipmaker’s 3nm process is still 10-20% lower compared to TSMC. As AI services expand in key markets such as mobile and server, chip power efficiency has emerged as one of the most critical factors in chipmaking.
“The main reason large customers choose TSMC is the difference in power efficiency between the companies’ chips at the front-end of the process,” said an official from a major contract chipmaker. “TSMC has increased the cost of producing 3nm chips by more than 25% compared to 5nm chips, making it more expensive than Samsung, but the performance difference drives customer preference for TSMC.”
Heat management is another challenge. “Heat generation in semiconductors has been a challenge for cutting-edge chipmakers over the past 20 years and has become the most important prerequisite in the era of AI chips,” the official added. “For mobile chips, excessive heat can collapse a smartphone’s structure, and for server chips, heat generated from a single server rack can spread like wildfire and overload the entire server system.”
The market share gap between TSMC and Samsung’s foundry business is also widening. According to market research firm TrendForce, Samsung Electronics’ foundry market share was 11% in the first quarter of this year, down from 11.3% in the fourth quarter of last year. TSMC’s market share rose from 61.2% to 61.7% during the same period despite declining smartphone demand.